Statistical methods for composite endpoints.

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Evidence Synthesis

Background:

• Composite endpoints are frequently used in clinical trials.
• Traditional time-to-first-event analysis assumes equal component severity and is sensitive to short-term events.
• Limitations of standard methods necessitate advanced statistical approaches.

Purpose of the Study:

• To review and compare novel statistical methods for analyzing composite endpoints.
• To highlight methods that address limitations of time-to-first-event analysis, particularly regarding event severity and frequency.

Main Methods:

• Review of advanced statistical methods including win ratio analysis, competing risk regression, negative binomial regression, Andersen-Gill regression, and weighted composite endpoint (WCE) analysis.
• Evaluation of advantages and limitations of each method.
• Focus on methods incorporating event severity and all patient events.

Main Results:

• Win ratio and WCE analyses account for event severity through pre-specified weights.
• Negative binomial and Andersen-Gill regression analyze all events per patient, potentially increasing statistical power.
• Each method presents unique advantages and limitations influencing treatment effect estimates.

Conclusions:

• Novel statistical methods can enhance the understanding of novel therapies, especially when composite endpoint components vary in severity and timing.
• Pre-specified methods are crucial for accurate treatment effect estimation.
• Consideration of patient types, drugs, devices, events, and follow-up duration is vital for appropriate method selection.